README.md

MMDIR

MMDIR: Multimodal Instruction-Driven Framework for Mixed-Degradation Document Image Restoration, CVPR 2026.

Proposed Benchmark (MixedDoc)

we introduce a novel benchmark named MixedDoc comprising complex mixed degradations, where each image contains randomized combinations of four degradation types (blur, shadow, text watermark, and seal).

The benchmark can be downloaded from Baidu Cloud.

The predict results by our model can be downloaded from Baidu Cloud.

License

This work need to be referenced under CC BY-NC-ND 4.0 License for non-commercial research purposes.

Acknowledgement

Thanks to M6Doc for their outstanding work in open-sourcing the original document images.

Citation

If our work is helpful to you, please refer to the following BibTeX format for citation:

@inproceedings{li2026mmdir,
  title={MMDIR: Multimodal Instruction-Driven Framework for Mixed-Degradation Document Image Restoration},
  author={Li, Heng and Wang, Xingyuan and Fan, Yang and Zhang, Yunan and Wu, Xiangping and Chen, Qingcai},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={8387--8396},
  year={2026}
}